AI regulation: enforce present law and assign accountability

Every few years a government decides a general-purpose technology is too important to leave alone, and writes rules for what it imagines the technology to be. Encryption is the example I keep coming back to. This week Steven Sinofsky retold that history as a parable for AI: a long run of attempts to control a core technology, followed by abuse of the controls themselves.

The lesson for executives is not that regulation is pointless. Rules written before anyone agrees on the failure modes tend to lock the wrong fix in place, and they are very hard to remove later.

Meanwhile the harms we have already seen, an agent reading material it had no right to read, a vendor that stayed quiet about it, sit inside law we already have. Computer misuse, negligence and disclosure duties do not need an AI amendment.

So I would put the next year of effort into two things. Enforce what exists against present harms. And write down, inside your own organisation, who answers for each agent's actions. Both are within an executive's reach this quarter. A new framework is not.

Source: Steven Sinofsky, Hardcore Software 243, 29 September 2026 https://lnkd.in/gJ2nqFdk